Visual feature adjustment-based light supplementing method for animation shooting

Abstract

The invention relates to the technical field of illumination control, in particular to a light supplementing method based on visual feature adjustment for animation shooting. The method comprises the following steps: controlling an addressable lighting array to perform multi-angle time-sharing sampling, and resolving a normal vector field of a scene surface; accurately dividing the high-brightness area into a low-frequency mirror reflection area and a high-frequency structure highlight area based on normal gradient distribution; performing brightness suppression on the low-frequency mirror reflection area, and performing neighborhood flux compensation based on a total luminous flux conservation principle; brightness maintenance or enhancement is performed on the high-frequency structured highlight region. According to the light supplementing method based on visual feature adjustment for animation shooting, interference oil light is eliminated, meanwhile, the structure texture is sharpened, and on the premise that the total illumination of a scene is kept stable, the light and shadow level and the reality sense of an animation picture are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of lighting control technology, and in particular to a supplementary lighting method based on visual feature adjustment for animation shooting. Background Technology

[0002] With the booming development of the digital creative industry, stop-motion animation, miniature models, and high-precision product still life photography have placed almost demanding requirements on the subtlety of light and shadow and the physical realism. In this specific field of artistic creation and industrial documentation, lighting not only serves the basic function of illuminating the subject, but is also a core means of depicting the texture of materials, shaping the sense of spatial structure, and conveying emotional atmosphere. To achieve a high level of visual presentation, shooting scenes often involve complex combinations of materials, such as clay, silicone, and fine resin. However, these materials have different physical surface properties, and under the illumination of traditional point or surface light sources, they are prone to producing high-intensity specular reflections at specific geometric transitions. These unintended strong reflective points often appear visually as an oily sheen that ruins the texture of the image or severe pixel-level overexposure noise. This not only obscures the color and details of the object itself, but also makes the carefully crafted model look cheap and plasticky, seriously affecting the artistic expression and image quality of the animation work.

[0003] To address the aforementioned issues, existing supplementary lighting technologies have differentiated and experimented with light sources from both physical and parametric perspectives at different historical stages. For example, the patent application with publication number CN118200506B employs a multi-angle mechanical supplementary lighting mechanism and its position adjustment system. Its purpose is to increase the physical redundancy of the light source in space and to change the shooting angle through mechanical displacement, thereby alleviating the blind spot problem caused by a fixed light source angle and significantly improving the efficiency of physical operations in the lighting process. Under certain historical technological conditions, this mechanical change based on hardware structure solved the problem of moving from immobility to multi-dimensional mobility.

[0004] Meanwhile, another technological evolution path explores software virtual mapping and parameter presets, such as the solution with publication number CN116188635A. This technology derives initial fill light parameters by associating them with animation shot files and optimizes these parameters using a user interface. The aim is to reduce reliance on the experience of senior lighting technicians through digital means, achieving preliminary proceduralization of lighting logic. This approach demonstrates good operability in virtual rendering or scenarios requiring relatively stable ambient light.

[0005] However, as shooting tasks demand higher precision in material reproduction and automation, the inherent limitations of the aforementioned technologies have become apparent. These limitations stem from a deep-seated and long-neglected technical contradiction: the contradiction between the lack of perceptual dimension and the blindness of light field adjustment. Whether it's mechanical adjustment of physical position or parameter mapping based on preset files, both are essentially open-loop control or global adjustment based on simple visual feedback. The mechanical adjustment scheme represented by CN118200506B relies entirely on preset paths or secondary intervention after manual observation. The system itself lacks real-time perception of the reflective characteristics and micro-geometry of the subject's surface. This means the system cannot physically distinguish which areas of highlight are structural highlights (used to enhance three-dimensionality) caused by the object's angular structure, and which areas are disruptive reflections (i.e., gloss) caused by smooth surfaces.

[0006] A deeper problem lies in the fact that existing brightness control strategies often fall into a trade-off between performance gains and losses. When existing systems detect overexposed areas in an image, they typically reduce the total power of the light source or adjust the global exposure. While this one-size-fits-all approach suppresses interfering reflections, it often sacrifices crucial shadow details and essential lighting at structural transitions due to a lack of understanding of the surface normal gradient. Fundamentally, this is because current technology has failed to establish a closed loop of "visual perception - geometric understanding - physical execution." In complex animation shooting conditions, the response of each pixel to light is determined by the direction of its surface normal, material albedo, and the angle of incident light. If the physical geometric characteristics of the object's surface cannot be obtained before or at the moment of shooting, any adjustment based on pixel brightness is merely a superficial parametric game, unable to accurately eliminate shine without sacrificing the object's texture. This gap in perception leads to unavoidable blindness in automated lighting, inevitably resulting in a flattened and distorted visual representation of objects while pursuing a shine-removing effect.

[0007] Therefore, how to reconstruct the geometric features of an object's surface using active visual perception technology in complex physical shooting environments, and how to achieve logically discernible differentiated modulation at the physical light field level based on this information, has become a critical technological barrier that urgently needs to be overcome in the field of automated animation shooting. How to accurately identify and suppress interfering reflections while maintaining total illuminance and energy balance, while preserving and enhancing structural highlights, has become a key challenge for improving the quality of digital creative shooting and achieving intelligent lighting decisions. Summary of the Invention

[0008] The purpose of this invention is to solve or at least mitigate the problems of inaccurate identification of interfering reflections on the surface of the subject caused by point light source illumination during animation shooting, as well as the flattening and loss of texture caused by existing adjustment methods, and to provide a supplementary lighting method based on visual feature adjustment for animation shooting.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a visual feature-based supplementary lighting method for animation shooting, applied to a shooting system including a camera and an addressable illumination array, comprising the following steps:

[0010] S1. Control the illumination array to perform multi-angle time-division illumination sampling on the scene to obtain at least three sampled images with different incident light angles;

[0011] S2. Based on the sampled image, the surface normal vector of each pixel in the scene is calculated using a photometric stereo algorithm to construct a surface normal vector field.

[0012] S3. Calculate the gradient distribution of the surface normal vector field, and classify the high-brightness areas in the scene into low-frequency specular reflection areas and high-frequency structural highlight areas based on the normal gradient features;

[0013] S4. Perform differentiated adjustment on the illumination array based on the differentiation results: suppress the brightness of the light-emitting units corresponding to the low-frequency specular reflection zone, and maintain or enhance the brightness of the light-emitting units corresponding to the high-frequency structural high-brightness zone.

[0014] To further realize the present invention, the following technical solutions may be preferred:

[0015] Preferably, before step S1, an initialization calibration step is also included:

[0016] The light-emitting units in the illumination array are illuminated traversally, and a calibration image is captured by the camera.

[0017] A light source-pixel mapping index table is constructed. The index table records the correspondence between pixel coordinates on the image plane and physical light-emitting units in the illumination array, as well as the illumination angle vector of each light-emitting unit for that pixel coordinate.

[0018] Preferably, step S1 specifically includes:

[0019] During the exposure interval or preview stage of the current video frame, at least three sets of non-coplanar light-emitting unit subsets are selected from the illumination array;

[0020] Each subset of light-emitting units is lit sequentially at microsecond intervals, and the camera is controlled to synchronously acquire the corresponding grayscale sampled images.

[0021] Preferably, step S2 further includes:

[0022] Based on the sampled image and the corresponding incident light angle, the surface albedo of each pixel in the scene is calculated.

[0023] The surface albedo is used as a weighting factor for adjusting the intensity of light in subsequent steps.

[0024] Preferably, step S3 specifically includes:

[0025] Differentiate the surface normal vector field to generate a normal gradient feature map;

[0026] Obtain the brightness distribution map of the current scene and extract the high-brightness areas;

[0027] Logically map the high-brightness region to the normal gradient feature map to divide the high-brightness region into a low-frequency specular reflection region and a high-frequency structural highlight region.

[0028] Preferably, the specific logic of the division is as follows:

[0029] If a pixel is located in a high-brightness area and its corresponding normal gradient value is less than a preset flatness threshold, it is determined to be a low-frequency specular reflection area.

[0030] If a pixel is located in a high-brightness area and its corresponding normal gradient value is greater than a preset texture threshold, it is determined to be a high-frequency structure highlight area.

[0031] Preferably, in step S4, the adjustment strategy for the low-frequency specular reflection region is as follows:

[0032] Query the light source-pixel mapping index table to determine the target light-emitting unit illuminating the area;

[0033] A negative gain control command is generated to reduce the driving current of the target light-emitting unit in order to eliminate specular reflection.

[0034] Preferably, when reducing the driving current of the target light-emitting unit, a neighborhood flux compensation strategy is implemented:

[0035] Identify neighboring light-emitting units of the target light-emitting unit in the physical space of the illumination array;

[0036] The brightness of the neighboring light-emitting units is increased so that the increased total luminous flux is equal to the reduced luminous flux of the target light-emitting unit, thereby achieving the conversion from hard light to diffuse light while maintaining a constant total illuminance in the local area.

[0037] Preferably, in step S4, the adjustment strategy for the high-frequency structure highlight region is as follows:

[0038] Maintain or increase the brightness of the target light-emitting unit illuminating the area;

[0039] Turn off or darken the light-emitting units around the target light-emitting unit to sharpen the light source boundary and enhance the three-dimensionality of the structure edge.

[0040] Preferably, the method further includes a closed-loop iteration step:

[0041] After performing step S4, the scene image is acquired again and the normal gradient features are calculated.

[0042] If the brightness of the low-frequency specular reflection area is still higher than the preset threshold, and the change in the normal gradient feature on the time axis is less than the hysteresis threshold, then repeat step S4 for fine-tuning.

[0043] The beneficial effects of this invention are:

[0044] This invention represents a leap in fill light logic, moving from brightness perception to geometric perception. By introducing active photometric stereo perception and normal gradient analysis, this invention overcomes the limitations of traditional fill light techniques, which can only adjust based on image brightness. For the first time, it achieves precise physical differentiation between low-frequency specular reflection areas (interfering reflections) and high-frequency structural highlight areas (structural highlights). This technological advancement ensures that while eliminating overexposure and shine, it perfectly preserves and enhances key light and shadow information that reflects the shape and structure of objects, completely resolving the technical pain point of a severely plastic-like appearance in images.

[0045] Meanwhile, this invention, based on a flux-conserving light field reshaping strategy, ensures the stability of scene illumination and image detail. The neighborhood flux compensation technology proposed in this invention does not simply shut down interfering light sources, but rather transforms direct, hard light into diffused, soft light of equal energy through precise energy transfer. This not only eliminates specular reflection but also avoids the problems of image darkening or loss of shadows caused by traditional local dimming, allowing the subject to maintain high brightness while exhibiting a more delicate and realistic physical texture.

[0046] This invention constructs a closed-loop physical control system of "perception-decision-execution". Through closed-loop control of "visual perception, geometric classification and physical compensation", it realizes refined and intelligent management of the lighting process of animation shooting, and solves the technical contradiction between highlight interference and material reproduction from the underlying physical logic. Attached Figure Description

[0047] Figure 1 This is the overall flowchart of the present invention.

[0048] Figure 2 This is a schematic diagram of the illumination semantic segmentation process based on normal gradient of the present invention.

[0049] Figure 3This is a schematic diagram of the LED array current distribution topology and neighborhood flux compensation principle of the present invention.

[0050] Figure 4 This is a schematic diagram of the optical path physical transmission model and optical quality reshaping of the present invention.

[0051] Figure 5 These are comparison images of the final imaging effects of the hybrid material character under different lighting strategies according to the present invention. Detailed Implementation

[0052] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1

[0055] In the current transformation of the digital creative industry towards refinement and automation, the granular requirements for lighting control in animation shooting (especially stop-motion animation and high-precision miniature model photography) have evolved from macroscopic lighting adjustments to microscopic light field reshaping. This implementation provides a visual feature-based supplementary lighting method for animation shooting, such as... Figure 1 As shown, the core technology lies in constructing a closed-loop optical control system that integrates active photometric stereo perception and physical luminous flux conservation modulation. This system is not simply a combination of image processing algorithms and lighting hardware; rather, it transforms the lighting array into an intelligent photon generator with both sensing and execution attributes. Through high-frequency sampling and semantic analysis of the geometric normal field of the object's surface, it reconstructs the energy distribution of the incident light field in real time in physical space, thereby resolving the contradiction between material reflection and structural texture representation at its source.

[0056] I. System Physical Architecture and Photoelectric Response Fundamentals

[0057] This implementation operates in a highly integrated optoelectronic physical environment. The core hardware architecture consists of a high-density addressable illumination array, an image acquisition device that is strictly time-synchronized with the illumination system, and a central computing unit responsible for computation and decision-making.

[0058] The illumination array constitutes the photon emission end of the system. Physically, it is a matrix-like planar or curved structure composed of thousands of independent light-emitting units, covering the forward hemisphere of the subject. Each light-emitting unit is an electrically independent control node, and the driving circuit of each unit is a high-bit-depth digital control. Unlike ordinary lights, each pixel in each light chain is an independent pulse width modulation (PWM) control channel. To ensure precise energy redistribution during subsequent neighborhood flux compensation, the PWM modulation frequency can reach the megahertz level, while the dimming resolution is 16 bits. This allows for precise control of the output power of any light-emitting unit at the microampere level, achieving a continuous and smooth conservation of light intensity. This is the foundation for light energy conservation in degreasing operations, because only with extremely high linearity and controllability can the reduced light energy be calculated, and only then can this reduced light energy be converted into an increase in neighboring units without loss.

[0059] The image acquisition device constitutes the photon receiver of the system. To support active photometric stereo sampling, an industrial camera with a global shutter is selected. The physical characteristics of the global shutter ensure that all pixels begin and end exposure at the same microsecond interval, thus eliminating the stripe artifacts produced by rolling shutters under nanosecond-level high-speed strobe illumination. The sensor and illumination array are connected by a hardware-triggered signal line, forming a nanosecond-level phase-locked loop. When the illumination array switches illumination modes, the sensor can enter integration mode with zero delay. This strict timing matching ensures that each frame of acquired image data is a pure record of this specific light field distribution.

[0060] The central computing unit (CCU) is the nerve center for perception and execution. Because the method involves a large amount of pixel-level normal calculation and real-time light field mask generation, the CCU is internally deployed with an architecture optimized for matrix operations. All light path mapping, gradient analysis, and energy conservation are processed at high speed in memory in the form of floating-point numbers. From light sampling to the issuance of supplementary lighting commands, the latency of the entire chain is lower than the human visual persistence threshold, thus providing animators with a real-time experience where what they think is what they see, and what they see is what they get.

[0061] II. Construction and Initialization of Spatial Mapping Relationships

[0062] A light source-pixel mapping index table is constructed. Any fine control must rely on a fine coordinate system. Therefore, before the system actually starts shooting, there is a necessary initialization calibration process, the purpose of which is to establish the equivalent and opposite correspondence between physical space and digital space.

[0063] The calibration logic is iterative. The control system drives each light-emitting unit in the lighting array to light up sequentially. In this process, the system doesn't simply record which light is on, but rather captures the luminous response of that light source in the subject scene using an image acquisition device. When the i-th light-emitting unit lights up, a set of pixels on the image plane... The brightness of the image undergoes a transition, and the intensity of this transition and the distribution of affected pixels are recorded. The physical meaning of this data is the transfer function that originates from a physical light source, propagates through space, is reflected by objects, and reaches the camera's image sensor.

[0064] By scanning the entire array, the system establishes a massive sparse matrix (i.e., an index table) in memory. This index table has two index dimensions: one index points from the index address of an LED to a corresponding index entry, and the other index points from the index address of a point on a 2D plane to an index entry. The first index reveals which pixels in the scene are illuminated by any given LED, along with their corresponding luminous intensity weights; the second index reveals which LEDs contribute to the brightness of any given pixel. Each entry in the index table contains the incident light angle vector corresponding to the LED on the 2D plane. This vector forms the geometric basis for subsequent photometric stereo normal calculations. The system calculates an accurate unit incident vector L, i.e., the unit incident light angle vector, using the physical coordinates of the emitting units in the array and the viewing direction of the pixel's back projection into 3D space. During this process, the system also performs effective optical path filtering, automatically identifying and eliminating invalid optical paths that prevent light from reaching the target pixel due to self-occlusion or obstacles, thereby improving the energy efficiency of subsequent calculations and securing greater spatial and radiometric resolution for photometric stereo normal calculations.

[0065] III. Active photometric stereo sampling and normal field inversion

[0066] After initialization, the system enters normal acquisition mode. Before each image is actually exposed to the subject, the system performs a time-division multi-view illumination sampling step. This is a key step in acquiring the geometric information of the subject, essentially using the illumination array temporarily as a high-precision 3D scanner.

[0067] This step is triggered based on strict timing control. The control system selects at least three subsets of luminescent units in non-coplanar spatial locations within the illumination array. These three sets of luminescent units are rapidly and alternately illuminated at microsecond intervals. Due to the high switching frequency of the illumination, what the naked eye sees is a faint, constant illumination of the array, but to the synchronized high-speed camera, it appears as three distinct black-and-white grayscale images that are strictly time-synchronized. These three images serve as the initial samples for photometric stereoscopic calculation. For each pixel in each image, the system now possesses observation data from three perspectives. Based on the Lambert reflection principle and its approximation for microsurfaces, brightness is a function of the angle between the incident light angle and the surface normal. The system establishes a system of equations simultaneously for each pixel, based on the three perspective observation equations.

[0068] Solving this system of equations is essentially an inversion of the microscopic geometry of the physical object's surface. The system obtains a three-dimensional vector *n* through mathematical calculation; this is the unit vector of the surface normal corresponding to that pixel, an objective, physically-based three-dimensional vector describing the surface orientation at that point. The normal vectors of all pixels in the entire image converge to form a high-precision surface normal vector field. This calculation process also produces an important byproduct: surface albedo. By calculating the ratio of geometric factors to brightness, the remaining value represents the original color and reflectivity of the object's surface. Albedo data will serve as an important weighting term in the subsequent supplementary lighting energy control, ensuring consistency in supplementary lighting treatment for dark and light materials.

[0069] IV. Illumination Semantic Segmentation and Visualization Analysis Based on Normal Gradient

[0070] With the normal vector field, the system gains the ability to understand the surface structure of physical objects. The fundamental difference between this method and simple brightness threshold lighting lies in segmenting and discriminating the geometric data according to illumination semantics, thereby achieving visualization and analysis: the system no longer focuses solely on where the light is bright, but rather on why it is bright. The main analytical tool is the normal gradient. The system differentiates the normal vector field, calculating the degree of difference in the normal direction between each pixel and its neighboring pixels, thereby generating a normal gradient heatmap.

[0071] like Figure 2 As shown, physically flat regions (such as a smooth sphere surface) are represented by a deep cool color (dark blue-black), with a gradient value of almost zero and a smooth, continuous gradient in the normal direction; physically rough regions (such as a rock surface or a fabric surface) are represented by a bright warm color (red-yellow), with a gradient value of huge jumps and a chaotic normal direction.

[0072] Based on the distribution patterns of heatmaps, binary logic is introduced to classify high-brightness areas (overexposed areas) on the heatmap:

[0073] In the low-frequency specular reflection region, the brightness exceeds the highlight threshold, but its normal gradient value falls within the cool color range of the heatmap (below the flatness threshold). In the physical context of animation shooting, this combination points to a specific physical phenomenon: a light source produces specular reflection on smooth, polished materials (varnished clay, the forehead of a plastic doll). This highlight lacks information; it's a glaring white patch, also known as shine. For visual arts, this is an interference that needs to be suppressed.

[0074] In the high-frequency structural highlight area, the brightness also exceeds the highlight threshold, but its normal gradient value is in the warm color range of the heatmap (above the texture threshold). The physical situation corresponding to this combination is that the light source hits the corner of an object, the brushed texture of metal, or the edge of a character's eyelid. This area is also very bright, but this brightness is caused by abrupt changes in geometric structure, carrying the object's volume, sharpness, and detail. For visual art, this is the texture of the object to be represented; it not only cannot be eliminated, but also needs to be preserved and even enhanced.

[0075] By generating optical field modulation masks based on different geometric physical features, these modulation masks are no longer simple binary maps, but instruction maps with semantic information, which tell the subsequent distribution of optical flux.

[0076] V. Differentiated optical field reshaping and topological current distribution

[0077] Based on the generated modulation mask, the system enters the light field reshaping stage, which transforms the digital decision into a redistribution of energy at the photonic level. This process is particularly evident in the current distribution topology of the LED array.

[0078] For the light path marked as a low-frequency specular reflection region, the system performs suppression and compensation operations. The control system first locks onto the target unit directly illuminating this region and significantly reduces its driving current. If it only stopped there, dark spots would appear on the surface of the object being photographed. Therefore, the system simultaneously activates a neighborhood flux compensation mechanism. This is an algorithm based on the law of conservation of energy, which forms a distinctive crater-like distribution pattern on the current topology of the LED array.

[0079] like Figure 3As shown, the left side illustrates the current distribution of a traditional point light source, with only the central unit outputting full power. The right side, however, displays the crater-like distribution processed by this invention, where the current in the central target unit is significantly depressed (suppressed), while the current in the neighboring units closely surrounding the center exhibits a ring-shaped bulge (enhanced). The mathematical relationships marked in the figure clearly show that the energy reduction at the center (e.g., 70%) is precisely and equally distributed to the increased energy in the surrounding areas (e.g., an 8.75% increase in each of the eight units), thus achieving the conservation of total energy. This crater-like current distribution produces profound optical effects in physical space, fundamentally altering its physical transport model.

[0080] like Figure 4 As shown in the image above, the optical path change in the suppression channel is illustrated: the narrow beam of hard light, originally concentrated at a narrow angle and causing strong specular reflection, is physically dispersed into a wide beam of diffused light entering from multiple surrounding angles with the total energy remaining constant. The solid angle of the incident light is significantly expanded, causing the reflected light to no longer concentrate and enter the lens, but rather to be evenly scattered in a soft, diffused manner. On the surface of the subject, this change manifests as the disappearance of glaring highlights, but the area does not darken; it's as if an invisible softbox has been applied.

[0081] On the other hand, for the optical path marked as a high-frequency structural highlight region, the system performs enhancement and sharpening operations. The control system locks onto the target unit illuminating this region, maintaining or finely adjusting its brightness. To further highlight the sharpness of the structure, the system performs a reverse sharpening operation: actively turning off or significantly darkening the surrounding units adjacent to the target unit. For example... Figure 4 As shown in the image below, this operation cuts off stray interfering light rays from the surrounding area, restoring the light source to an approximate collimated point light source. Point light sources have extremely strong directionality and shadow casting capabilities, and can sculpt the undulations and textures of object edges like a carving knife, making structural highlights crisper and cleaner.

[0082] VI. Closed-loop verification and system stability

[0083] After the light field reshaping command is issued, the system enters the closed-loop verification phase. The image acquisition device recaptures the image, and the calculation unit extracts the residual brightness of the original low-frequency specular reflection area. If the brightness is still higher than the diffuse reflection standard value, the system generates a correction coefficient for iterative fine-tuning. This negative feedback mechanism ensures the system's adaptive convergence capability for materials with different reflectivities.

[0084] To prevent light flickering caused by sensor thermal noise or micrometer-level displacement during continuous shooting, the system employs a temporal hysteresis filtering mechanism. The system maintains a historical queue of normal gradient features in memory. Only when the detected geometric feature displacement exceeds a preset stability threshold does the system consider a substantial change in the scene, thus triggering the calculation of new lighting parameters. Conversely, if the change is within the threshold, the system forcibly locks the current LED drive parameters to ensure absolutely constant illumination output.

[0085] Example 2

[0086] To quantitatively verify the actual effect of the method of the present invention, a representative high-difficulty shooting scene was constructed in a standard stop-motion animation studio environment, and a detailed comparative experiment was conducted.

[0087] Experimental scenario setup:

[0088] The subject is a stop-motion animated character made of mixed materials. The character's head is made of resin clay coated with varnish, with a smooth surface and continuous curvature, making it prone to specular reflection (gloss). The character's shoulders are covered in rough linen clothing with a rich surface texture, which relies on high-frequency lighting to represent the texture. This combination of a bald head and linen clothing is a nightmare for traditional lighting, because soft lighting on the clothing will sacrifice the hair's natural shine, while hard lighting on the clothing will cause the hair to become excessively oily.

[0089] Comparative example (traditional technique): Automatic exposure (adjustment) based on a global brightness threshold. If an overexposed area (brightness greater than 235) is detected in an image, all light sources associated with that overexposed area are uniformly reduced in brightness, i.e., geometric features are not considered and there is no neighborhood compensation.

[0090] Reference example (in this invention): Enable a full set of visual feature adjustments, namely normal gradient segmentation and flux conservation compensation.

[0091] Visual effects comparison and analysis:

[0092] like Figure 5 As shown, significant visual differences were discovered in the experimental imaging results, which coincide with the aforementioned crater current and optical path model:

[0093] 1. In the head region (low-frequency specular reflection area):

[0094] Left image (comparative): To eliminate highlight clipping on the forehead, traditional systems can only turn the main light very low. The result is that the highlight is not clipped, but the entire face turns gray, the skin tone lacks translucency, and the nature of the light source has not changed, so the highlight is still harsh, just darkened.

[0095] Right image (Example): The glaring spot on the forehead has completely disappeared, replaced by a laminated, soft, and uniform diffused glow. Importantly, the brightness of the face has not decreased, nor has the skin tone darkened. This is the most direct evidence that neighborhood flux compensation has diffused direct light.

[0096] 2. In the shoulder area (high-frequency structural highlight area):

[0097] Left image (comparative): Due to the global light reduction, the linen texture, which originally needed strong light to outline, became blurred, the three-dimensionality of the warp and weft lines was smoothed out, and the image appeared flat.

[0098] Right image (Example): The texture of the linen clothing is not weakened; on the contrary, it is clearer than before. Each fiber casts a slight shadow under the illumination of a pinpoint light source, resulting in excellent texture. This is the result of a sharpening modulation strategy that preserves and enhances high-frequency details.

[0099] To more objectively evaluate the technical effect, we collected photometric data for key areas, as shown in the table below:

[0100] Key performance indicators Comparative example (traditional global threshold dimming) Reference Example (Visual Feature Adjustment of the Invention) Performance Improvement and Analysis Peak brightness of the specular reflection area 248 Critical Overflow 175 Ideal diffuse reflection range Effective suppression: Successfully eliminated the risk of overexposure, and the values ​​are more consistent with the human eye's perception of diffuse reflection from skin. Average illuminance deviation in the specular reflection area -42% Severe underexposure -1.5% basically constant Flux conservation: This demonstrates that the neighborhood compensation mechanism of this invention successfully maintains the total energy of the region, avoiding the common problem of the oil turning black immediately after removal of shine. Structural highlight texture contrast 1.15 : 1 4.60 : 1 Enhanced texture: The contrast ratio was improved by more than 3 times, indicating that the system successfully identified and sharpened texture details through geometric semantic segmentation. Shadow detail retention 55% 98% Dynamic range: Because there is no need to globally darken the image to suppress highlights, shadow details are fully preserved.

[0101] The above comparison demonstrates that the method described in this embodiment has an overwhelming advantage when processing complex mixed materials. By reconstructing the light field at the physical level, it overcomes the physical-optical contradictions that cannot be resolved by simple image processing or simple dimming, achieving the goal of simultaneously softening smooth surfaces and sharpening rough surfaces in the same image.

[0102] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A visual feature-based supplementary lighting method for animation shooting, applied to a shooting system including a camera and an addressable illumination array, characterized in that, Includes the following steps: S1. Control the illumination array to perform multi-angle time-division illumination sampling on the scene to obtain at least three sampled images with different incident light angles; S2. Based on the sampled image, the surface normal vector of each pixel in the scene is calculated using a photometric stereo algorithm to construct a surface normal vector field. S3. Calculate the gradient distribution of the surface normal vector field, and classify the high-brightness areas in the scene into low-frequency specular reflection areas and high-frequency structural highlight areas based on the normal gradient features; S4. Perform differentiated adjustment on the illumination array based on the differentiation results: suppress the brightness of the light-emitting units corresponding to the low-frequency specular reflection zone, and maintain or enhance the brightness of the light-emitting units corresponding to the high-frequency structural high-brightness zone.

2. The method according to claim 1, characterized in that, Before step S1, an initialization calibration step is also included: The light-emitting units in the illumination array are illuminated traversally, and a calibration image is captured by the camera. A light source-pixel mapping index table is constructed. The index table records the correspondence between pixel coordinates on the image plane and physical light-emitting units in the illumination array, as well as the illumination angle vector of each light-emitting unit for that pixel coordinate.

3. The method according to claim 1, characterized in that, Step S1 specifically includes: During the exposure interval or preview stage of the current video frame, at least three sets of non-coplanar light-emitting unit subsets are selected from the illumination array; Each subset of light-emitting units is lit sequentially at microsecond intervals, and the camera is controlled to synchronously acquire the corresponding grayscale sampled images.

4. The method according to claim 1, characterized in that, Step S2 further includes: Based on the sampled image and the corresponding incident light angle, the surface albedo of each pixel in the scene is calculated. The surface albedo is used as a weighting factor for adjusting the intensity of light in subsequent steps.

5. The method according to claim 1, characterized in that, Step S3 specifically includes: Differentiate the surface normal vector field to generate a normal gradient feature map; Obtain the brightness distribution map of the current scene and extract the high-brightness areas; Logically map the high-brightness region to the normal gradient feature map to divide the high-brightness region into a low-frequency specular reflection region and a high-frequency structural highlight region.

6. The method according to claim 5, characterized in that, The specific logic of the division is as follows: If a pixel is located in a high-brightness area and its corresponding normal gradient value is less than a preset flatness threshold, it is determined to be a low-frequency specular reflection area. If a pixel is located in a high-brightness area and its corresponding normal gradient value is greater than a preset texture threshold, it is determined to be a high-frequency structure highlight area.

7. The method according to claim 6, characterized in that, In step S4, the adjustment strategy for the low-frequency specular reflection region is as follows: Query the light source-pixel mapping index table to determine the target light-emitting unit illuminating the area; A negative gain control command is generated to reduce the driving current of the target light-emitting unit in order to eliminate specular reflection.

8. The method according to claim 7, characterized in that, When reducing the driving current of the target light-emitting unit, a neighborhood flux compensation strategy is executed: Identify neighboring light-emitting units of the target light-emitting unit in the physical space of the illumination array; The brightness of the neighboring light-emitting units is increased so that the increased total luminous flux is equal to the reduced luminous flux of the target light-emitting unit, thereby achieving the conversion from hard light to diffuse light while maintaining a constant total illuminance in the local area.

9. The method according to claim 6, characterized in that, In step S4, the adjustment strategy for the high-frequency structure high-brightness region is as follows: Maintain or increase the brightness of the target light-emitting unit illuminating the area; Turn off or darken the light-emitting units around the target light-emitting unit to sharpen the light source boundary and enhance the three-dimensionality of the structure edge.

10. The method according to any one of claims 1 to 9, characterized in that, The method also includes a closed-loop iterative step: After performing step S4, the scene image is acquired again and the normal gradient features are calculated. If the brightness of the low-frequency specular reflection area is still higher than the preset threshold, and the change in the normal gradient feature on the time axis is less than the hysteresis threshold, then repeat step S4 for fine-tuning.

Images